TW202013104A - Data processing method, data processing device, and computer-readable recording medium - Google Patents

Abstract

A data processing method that processes a plurality of unit processing data (each unit processing data include plural types of time-series data) includes an evaluation value distribution utilization step, in which processing that uses evaluation value distributions showing degrees of each value of evaluation values obtained by evaluating each time-series datum is carried out (for example, a step in which each time-series datum is compared with reference data and scoring that quantifies results obtained thereby as the evaluation values is carried out, and a step in which judgment of abnormality degrees is carried out using the evaluation value distributions based on results of the scoring); and an evaluation value distribution update step, in which the evaluation value distributions are updated.

Description

Data processing, data processing device and computer readable recording medium

The invention relates to a digital data (digital data) processing, in particular to a method for processing time series data.

As a method for detecting an abnormality of a device or device, the following method is known: a sensor or the like is used to measure a physical quantity (such as length, angle, time, speed, force, and pressure) indicating the operating state of the device or device , Voltage, current, temperature, flow, etc.), and analyze the time series data obtained by arranging the measurement results in the order of occurrence. When the equipment or device performs the same action under the same conditions, if there is no abnormality, the time series data changes similarly. Therefore, by comparing the same time-series data with each other to detect abnormal time-series data, and analyzing the abnormal time-series data, the location of the abnormality or the cause of the abnormality can be determined. Moreover, in recent years, the data processing capabilities of computers have improved significantly. Therefore, even if the amount of data is huge, it is often possible to obtain the desired result in a practical time. Therefore, the analysis of time series data has gradually become popular.

For example, in the field of semiconductor substrate manufacturing, the analysis of time series data has gradually become popular. In the manufacturing process of a semiconductor substrate (hereinafter referred to as a “substrate”), a series of processes are performed by a substrate processing device. The substrate processing apparatus includes a plurality of processing units that perform a specific process in a series of processes on the substrate. Each processing unit processes the substrate according to a predetermined flow (referred to as "recipe"). At this time, based on the measurement results in each processing unit, time series data is obtained. By analyzing the obtained time series data more often, it is possible to determine the processing unit or the cause of the abnormality. In addition, the term "recipe" is used not only for the processing of the substrate, but also for pre-processing before the substrate processing, or for maintaining the state of the processing unit while the processing unit is not processing the substrate/ Management or processing of various measurements related to the processing unit. However, in this specification, attention is paid to the processing performed on the substrate. In addition, Japanese Patent Laid-Open No. 2017-83985 discloses an invention related to the calculation of the abnormality of time-series data obtained by manufacturing a substrate.

In general, in the manufacturing process of the substrate, through the execution of the recipe, time series data on a large number of parameters (various physical quantities) is obtained. The time-series data is data that uses sensors to measure various physical quantities (for example, the flow rate or temperature of the processing fluid supplied from the nozzle (nozzle), the temperature in the chamber (chamber) Humidity, the internal pressure of the chamber, the exhaust pressure of the chamber, etc.), and the measurement results are arranged in accordance with the time series. In addition, the data obtained by analyzing the images captured by the camera (camera) in time series also becomes time series data. In addition, the determination of whether each time series data is abnormal is made by comparing the data value of the time series data with a threshold, or the value calculated from the data value according to a prescribed calculation rule (rule) Compare with threshold. In addition, the threshold is set for each parameter.

However, since the set threshold is not necessarily an ideal value, the accuracy of abnormality determination is not good. That is, according to the conventional method, it is impossible to accurately detect the abnormality of the time series data. Moreover, even when looking at a certain same recipe, the content of the obtained time series data tends to change with the passage of time. Therefore, if such time lapse is not taken into consideration in the abnormality determination, the abnormality cannot be detected with sufficient accuracy.

Therefore, an object of the present invention is to provide a data processing method capable of performing abnormality detection using time-series data with sufficient accuracy in consideration of the passage of time.

One aspect of the present invention is a data processing method that uses multiple time series data obtained by unit processing as unit processing data and processes multiple unit processing data. The data processing method includes: an evaluation value distribution utilization step, A process using an evaluation value distribution that represents the degree of each value of the evaluation value obtained by evaluating each time series data is performed; and an evaluation value distribution update step that updates the evaluation value distribution.

According to such a configuration, a process using an evaluation value distribution that represents the distribution of evaluation values obtained by evaluating each time series data is performed. For example, when time-series data is newly obtained, abnormality determination of the time-series data can be performed. When the evaluation value distribution is used in the abnormality determination, the evaluation value distribution is updated. Therefore, in the abnormality determination, for example, the latest trend of time series data can be considered. Based on the above, it is possible to perform abnormality detection using time series data with sufficient accuracy in consideration of the passage of time.

Another aspect of the present invention is a data processing device that uses multiple time series data obtained by unit processing as unit processing data to process multiple unit processing data. The data processing device includes: an evaluation value distribution utilization unit , A process using an evaluation value distribution, the evaluation value distribution representing the degree of each value of the evaluation value obtained by evaluating each time series data; and an evaluation value distribution update unit that updates the evaluation value distribution.

Another aspect of the present invention is a computer-readable recording medium that stores a data processing program that is used to cause a computer to perform an evaluation value distribution utilization step and an evaluation value distribution update step. The obtained multiple time series data are used as unit processing data to process the multiple unit processing data: the evaluation value distribution utilizes the step of processing using the evaluation value distribution, the evaluation value distribution indicates that it is obtained by evaluating each time series data The degree of each value of the evaluation value, the evaluation value distribution updating step updates the evaluation value distribution.

The above and other objects, features, forms, and effects of the present invention will be further clarified from the following detailed description of the present invention with reference to the drawings.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

＜1. Overall structure＞ 1 is a block diagram showing the overall configuration of a data processing system (data processing system for a substrate processing apparatus) according to an embodiment of the present invention. The data processing system includes a data processing device 100 and a substrate processing device 200. The data processing apparatus 100 and the substrate processing apparatus 200 are connected to each other through a communication line 300.

The data processing device 100 functionally includes a unit processing data selection unit 110, an evaluation value calculation unit 120, an evaluation value distribution creation unit 130, an evaluation value distribution update unit 140, an abnormality determination unit 150, and a data storage unit 160. The unit processing data selection unit 110 selects two or more unit processing data from a plurality of stored unit processing data to be described later. The evaluation value calculation unit 120 performs calculation of an evaluation value used for determination of abnormality of time-series data obtained by substrate processing, and the like. For example, the evaluation value calculation unit 120 calculates an evaluation value for each time series data included in the unit processing data selected by the unit processing data selection unit 110. The evaluation value distribution creation unit 130 creates an evaluation value distribution described later based on the evaluation value calculated by the evaluation value calculation unit 120 (evaluation value for each time series data). The evaluation value distribution update unit 140 updates the evaluation value distribution. The abnormality determination unit 150 determines the abnormality of the time series data newly obtained by executing the recipe by the substrate processing apparatus 200 based on the evaluation value and the evaluation value distribution of the time series data when the evaluation value distribution already exists . The evaluation value distribution utilization unit is realized by the evaluation value calculation unit 120 and the abnormality determination unit 150. In addition, in this embodiment, as a result of substrate processing, it is assumed that the smaller the evaluation value, the better.

The data storage unit 160 holds a data processing program 161 for executing various processes in this embodiment. The data storage unit 160 includes a time-series data DB 162 storing time-series data transmitted from the substrate processing apparatus 200, a reference data DB 163 storing reference data, and an evaluation value distribution data DB 164 storing evaluation value distribution data. The description of the reference data and the evaluation value distribution data will be described later. In addition, "DB" is an abbreviation for "database".

The substrate processing apparatus 200 includes a plurality of processing units 222. In each processing unit 222, a plurality of physical quantities indicating the operating state of the processing unit 222 are measured. Thus, multiple time series data (more specifically, time series data on multiple parameters) are obtained. The time series data obtained by the processing in each processing unit 222 is sent from the substrate processing apparatus 200 to the data processing apparatus 100, and is stored in the time series data DB 162 as described above.

FIG. 2 is a diagram showing a schematic structure of the substrate processing apparatus 200. The substrate processing apparatus 200 includes an indexer unit 210 and a processing unit 220. The control of the positioner unit 210 and the processing unit 220 is performed by a control unit (not shown) inside the substrate processing apparatus 200.

The positioner unit 210 includes: a plurality of substrate container holding parts 212 for placing a substrate container (cassette) that can accommodate a plurality of substrates; and an indexer robot 214 that performs substrate removal from the substrate The carrying out of the receiving container and the carrying in of the substrate into the substrate receiving container. The processing unit 220 includes a plurality of processing units 222 that use a processing liquid to perform processing such as cleaning of the substrate; and a substrate transport robot 224 that carries the substrate into and out of the processing unit 222. The number of processing units 222 is, for example, twelve. At this time, for example, a tower structure formed by stacking three processing units 222 is provided at four places around the substrate transfer robot 224 as shown in FIG. 2. Each processing unit 222 is provided with a chamber that is a space for processing a substrate, and a processing liquid is supplied to the substrate in the chamber. In addition, each processing unit 222 includes one chamber. That is, the processing unit 222 and the chamber are in one-to-one correspondence.

When processing the substrate, the positioner robot 214 takes out the processing object substrate from the substrate container placed on the substrate container holding portion 212 and hands the substrate to the substrate transport robot 224 via the substrate transfer portion 230. The substrate transfer robot 224 transfers the substrate received from the positioner robot 214 to the object processing unit 222. When the processing of the substrate is completed, the substrate transfer robot 224 takes out the substrate from the object processing unit 222 and delivers the substrate to the positioner robot 214 via the substrate transfer section 230. The positioner robot 214 transfers the substrate received from the substrate transfer robot 224 to the target substrate container.

In the data processing system, in order to detect an equipment abnormality related to the processing in each processing unit 222 or an abnormality in processing performed by each processing unit 222, each time a recipe is executed, time series data is acquired. The time series data obtained in this embodiment is data as follows, that is, when performing a recipe, a sensor or the like is used to measure various physical quantities (such as the flow rate of the nozzle, the internal pressure of the chamber, and the exhaust pressure of the chamber Etc.), and arrange the obtained data according to the time series. Various physical quantities are treated as the values of corresponding parameters, respectively. In addition, one parameter corresponds to one physical quantity.

FIG. 3 is a graph showing a certain time series data. The time series data is data about a certain physical quantity obtained by processing a piece of substrate in a chamber in a processing unit 222 when executing a recipe. In addition, time series data is data containing multiple discrete values, but in Figure 3, two data values adjacent in time are connected by a straight line. In addition, when one recipe is executed, for each processing unit 222 that executes the recipe, time series data about various physical quantities is obtained. Therefore, in the following, when a recipe is executed, the processing performed on one substrate in a chamber in one processing unit 222 is referred to as "unit processing", and a group of time series data obtained by unit processing is referred to as "unit processing" data". In a unit processing data, as shown schematically in FIG. 4, it contains time series data and attribute data about multiple parameters, and the attribute data includes multiple items (such as processed Data such as start time, end time of processing, etc.). In addition, regarding FIG. 4, “parameter A”, “parameter B” and “parameter C” correspond to physical quantities of different types from each other.

In order to detect abnormalities in equipment or processing, the unit processing data obtained through the execution of the recipe should be compared with the unit processing data having the ideal data value as the processing result. In more detail, the multiple time series data contained in the unit processing data obtained by the execution of the recipe and the multiple time contained in the unit processing data with the ideal data value as the processing result Compare serial data. Therefore, in this embodiment, regarding each recipe, unit processing data (including multiple time series included in the unit processing data used as the evaluation object) for comparing with the unit processing data used as the evaluation object The unit processing data of multiple time series data for which the data are compared separately is set as the benchmark data (the data used as the benchmark when calculating the evaluation value). The reference material is stored in the reference material DB163 (refer to FIG. 1). In addition, regarding each recipe, the time series data contained in the processing data of different units can be used as the reference data for each parameter. That is, when focusing on the parameters associated with a recipe, the time series data processed as the reference data about a certain parameter and the time series data processed as the reference data about other parameters may also be different from each other. The unit processes the time series data contained in the data.

Here, the hardware structure of the data processing device 100 will be described with reference to FIG. 5. The data processing device 100 includes a central processing unit (Central Processing Unit, CPU) 11, a main memory 12, an auxiliary memory device 13, a display unit 14, an input unit 15, a communication control unit 16, and a recording medium reading unit 17. The CPU 11 performs various arithmetic processing and the like in accordance with the given command. The main memory 12 temporarily stores running programs or data. The auxiliary memory device 13 stores various programs and various data that should be maintained even if the power is turned off. The data storage unit 160 is realized by the auxiliary memory device 13. The display unit 14 displays various screens for an operator to perform work, for example. For the display unit 14, for example, a liquid crystal display (display) is used. The input unit 15 is, for example, a mouse, a keyboard, or the like, and accepts input from the outside by the operator. The communication control unit 16 controls data transmission and reception. The recording medium reading unit 17 is an interface circuit of the recording medium 400 on which programs and the like are recorded. For the recording medium 400, for example, a non-transitory recording medium such as a Compact Disc Read-Only Memory (CD-ROM) or a Digital Versatile Disc Read-Only Memory (DVD-ROM) is used.

When the data processing device 100 is started, the data processing program 161 (refer to FIG. 1) held by the auxiliary memory device 13 (data storage unit 160) is read into the main memory 12, and the CPU 11 executes reading to the main memory The data processing program 161 in the memory 12. Thus, the data processing device 100 provides functions for performing various data processing. In addition, the data processing program 161 is provided, for example, in the form of being recorded in the recording medium 400 such as a CD-ROM or DVD-ROM, or in the form of downloading via the communication line 300.

<2. Evaluation of substrate processing> <2.1 Evaluation Value Distribution> In the present embodiment, in order to perform abnormality determination on each time-series data, an evaluation value distribution is used, and the evaluation value distribution indicates the degree of each value of the evaluation value obtained by the evaluation value calculation unit 120. The distribution of the evaluation values will be described in detail with reference to FIG. 6.

The evaluation value distribution is prepared for each parameter (ie, each time series data). When focusing on a certain parameter, the distribution of the degrees representing each evaluation value of the time-series data becomes, for example, as shown in Part A of FIG. 6. Regarding part A of FIG. 6, μ is the average value of the evaluation values of the distribution generation source, and σ is the standard deviation of the evaluation values of the distribution generation source. Here, by normalizing the evaluation values of the distribution generation sources, the distribution shown in Part B of FIG. 6 (the distribution with the average value of 0 and the dispersion/standard deviation of 1) can be created as the evaluation value distribution 5. In addition, if the evaluation value before normalization is Sold and the evaluation value after normalization is Snew, normalization is performed by the following formula (1).

In the case where the evaluation value distribution 5 as described above is prepared, when time series data is newly generated by execution of the recipe, the evaluation value for the time series data is obtained. In addition, with respect to the obtained evaluation value, the average value μ and the standard deviation σ when the evaluation value distribution 5 is created are used to perform normalization based on the above formula (1). Based on the evaluation value obtained by the standardization, the abnormality of the corresponding time series data is determined.

Regarding the determination of the degree of abnormality, in the present embodiment, the range of the standardized evaluation value is divided into four zones. That is, the abnormality of each time-series data is determined at four levels. Specifically, as shown in part B of FIG. 6, if the evaluation value (after normalization) is less than 1, the abnormality level is determined to be level 1 (L1), and if the evaluation value is 1 or more and less than 2, the abnormality level is determined to be In level 2 (L2), if the evaluation value is 2 or more and less than 3, the abnormality level is determined to be level 3 (L3), and if the evaluation value is 3 or higher, the abnormality level is determined to be level 4 (L4).

In addition, the division of the four areas regarding the standardized evaluation value range is based on the standard deviation. That is, the threshold of the interval is automatically determined. Therefore, unlike in the past, the user does not need to set a tedious task of setting a threshold in order to determine abnormality of time series data.

<2.2 Overall processing flow> FIG. 7 is a flowchart showing an overview of the overall processing flow for abnormality detection using time-series data. In addition, it is assumed that a certain amount of time series data has been stored before the start of the processing.

First, in order to make it possible to detect abnormality using time-series data (abnormality determination on each time-series data), the evaluation value distribution 5 is created (step S10 ). In this embodiment, the evaluation value distribution 5 common to all processing units 222 is created for each parameter. However, the evaluation value distribution 5 for each parameter may be created for each processing unit 222. The detailed flow of the evaluation value distribution 5 will be described later.

Next, the user specifies the processing unit (chamber) as the abnormality determination object and the parameters (step S20). At this time, the display unit 14 of the data processing device 100 displays, for example, the abnormality determination object setting screen shown in FIG. 8 (in FIG. 8, only a part of the screen actually displayed is shown, and also in FIGS. 11, 12, and 13 Similarly) 500, the user specifies the processing unit and parameters to be set as the abnormality determination object. In the example shown in FIG. 8, the processing unit whose check box becomes the selected state and the parameter that becomes the selected state within the list box are designated as the object of abnormality determination. In addition, in step S10, the time series data obtained by the processing in all the processing units 222 is used to create evaluation value distributions 5 for all parameters, but only through the processing in the processing unit specified in step S20 Among the obtained time-series data, the time-series data regarding the parameter specified in step S20 becomes the object of actually performing abnormality determination.

Subsequently, when the recipe is executed by the substrate processing apparatus 200 (step S30), scoring is performed on the time-series data as an abnormality determination object among the time-series data obtained by execution of the recipe (step S40) ). In addition, the so-called grading refers to a process of comparing each time series data with reference data and quantifying the result obtained thereby into an evaluation value. After the scoring is completed, regarding each time series data, the corresponding evaluation value distribution 5 is used to determine the abnormality (step S50 ). In the step S50, first, the evaluation value obtained in the step S40 is standardized. The normalization of the evaluation value is performed by the above formula (1), and for the average value μ and the standard deviation σ in the above formula (1), the average value μ and the standard deviation obtained at the time of making the corresponding evaluation value distribution 5 are used σ. Then, the degree of abnormality is determined based on the position of the standardized evaluation value on the evaluation value distribution 5. For example, if the normalized evaluation value is the value at the position marked with a symbol 51 in FIG. 9, the abnormality of the corresponding time-series data is determined to be “level 2”.

In the present embodiment, the processing from step S30 to step S50 is repeated until there is a change in the content of any recipe. That is, the same evaluation value distribution 5 is used to determine the degree of abnormality when a certain recipe is executed until the content of the recipe changes. When there is a change in the content of any recipe, the evaluation value distribution 5 is updated (step S60). According to the present embodiment, since the evaluation value distribution is updated in this way, for example, abnormality detection using time series data can be performed in consideration of recent trends. In addition, the details of the update of the evaluation value distribution 5 will be described later. After the update of the evaluation value distribution 5, the process returns to step S30.

In addition, in the present embodiment, the evaluation value distribution utilization step is realized by steps S40 and S50, and the evaluation value distribution update step is realized by step S60.

<3. How to make the evaluation value distribution> The detailed flow of creating the evaluation value distribution 5 (step S10 in FIG. 7) in this embodiment will be described with reference to FIG. 10. First, the user selects two or more unit processing materials that are the source of the evaluation value distribution 5 (step S110). In step S110, the display unit 14 of the data processing device 100 displays, for example, the unit processing data selection screen 600 shown in FIG. The unit processing data selection screen 600 includes a start time point input box 61, an end time point input box 62, a processing unit designation box 63, a recipe designation box 64, an extracted data display area 65, and an OK button 66. The start time point input box 61 and the end time point input box 62 are list boxes that can specify a date and time, and the processing unit designation box 63 and the recipe designation box 64 are list boxes that can select more than one item from a plurality of items. The user executes the processing unit through the start time point input box 61 and the end time point input box 62, the processing unit through the processing unit designation box 63, and the recipe through the recipe designation box 64. As a result, a list of unit processing data corresponding to the specified conditions is displayed in the extracted data display area 65. The user presses the OK button 66 in a state where part or all of the unit processing data displayed in the extracted data display area 65 is selected. Thereby, the unit processing data that is the source of the evaluation value distribution 5 is specified. In addition, the period, the processing unit, and the recipe need not all be specified, as long as at least any one of the execution period, the processing unit, and the recipe is sufficient.

Next, the evaluation value is calculated for each time-series data included in the unit processing data selected in step S110 (hereinafter referred to as “selected unit processing data”) (step S111 ). In this embodiment, the reference data is held in the reference data DB 163 in advance. That is, the reference data to be compared with each time series data included in the selected unit processing data is held in the reference data DB163. Therefore, in step S111, each time series data included in the selected unit processing data is compared with the reference data held in the reference data DB 163 (see FIG. 1), and the evaluation value for the respective time series data is calculated .

Next, the evaluation value calculated in step S111 is normalized (step S112). As described above, the normalization of the evaluation value is performed using the above formula (1). In this case, the evaluation value distribution 5 is created for each parameter. Therefore, the average value μ and the standard deviation σ in the above formula (1) are obtained for each parameter.

Finally, for each parameter (ie, each type of time series data), an evaluation value distribution 5 is created based on the standardized evaluation value data (step S113). The data constituting the evaluation value distribution 5 is held as the evaluation value distribution data in the evaluation value distribution data DB164 (refer to FIG. 1 ).

<4. Update method of evaluation value distribution> Next, the update of the evaluation value distribution 5 will be described. The unit processing data obtained by executing the recipe by the substrate processing apparatus 200 includes time series data on a plurality of parameters. As described above, in the present embodiment, the evaluation value distribution 5 is created for each of the parameters. In addition, in the substrate processing apparatus 200, the content of the recipe may be changed. If there is a change in the content of the recipe, before and after the change, the content of the time series data obtained by the execution of the recipe will be different. At this time, if the evaluation value distribution 5 created before the recipe change is used to perform the abnormality determination of the time-series data obtained after the recipe change, there may be no accurate result as the result of the abnormality determination. Therefore, in the present embodiment, when there is a change in the content of the recipe, the evaluation value distribution 5 is updated. In addition, after the content of the recipe has changed, since the time series data based on the changed content has not been stored, it is preferable that the update of the evaluation value distribution 5 is to store a certain time series based on the changed content Information only.

When the evaluation value distribution 5 is updated, the evaluation value distribution update unit 140 compares the parameters related to the recipe before the change with the parameters related to the recipe after the change. Furthermore, the evaluation value distribution update unit 140 creates an evaluation value distribution 5 corresponding to the parameter added with the change of the recipe content based on the stored evaluation value (evaluation value of the time-series data of the corresponding parameter). Then, the user specifies the parameter whose content is changed, and the evaluation value distribution update unit 140 recreates the evaluation value distribution 5 corresponding to the specified parameter.

For example, suppose that the content of a certain recipe changes and the parameter group associated with the recipe changes as follows. Before change: parameter A, parameter B, parameter C, parameter D After the change: parameter A, parameter C, parameter D, parameter E In addition, it is assumed that there is no change in the content of the time series data regarding parameter A and parameter D, and there is a change in the content of the time series data regarding parameter C.

In the case of the above example, when the evaluation value distribution 5 is updated, for example, the parameter designation screen 700 shown in FIG. 12 is displayed on the display unit 14 of the material processing device 100. The parameter designation screen 700 includes check boxes corresponding to the changed parameter group (parameter A, parameter C, parameter D, parameter E). The check box corresponding to parameter E, which is a parameter added with the change of the content of the recipe, has been selected in advance (shaded state in FIG. 12). On this parameter designation screen 700, regarding the parameter C, since the content of the time series data is changed, as shown in FIG. 13, the user sets the check box corresponding to the parameter C to the selected state. In this way, after the user executes the parameter, the evaluation value distribution 5 is actually updated. As a result, as shown schematically in FIG. 14, the evaluation value distribution 5 is updated. Specifically, the evaluation value distribution 5 of parameter B, which is a parameter deleted with the change of the recipe content, is deleted, and the evaluation value distribution 5 of parameter E, which is a parameter added with the change of the recipe content, is newly created, and the The parameter designated by the user, that is, the evaluation value distribution of the parameter C5. In addition, the evaluation value distribution 5 for the parameters A and D is maintained in the state before the recipe content was changed.

As described above, only the evaluation value distribution 5 regarding the parameters related to the change of the recipe content is updated (created, recreated, deleted). Thus, it takes a huge time to prevent the update of the evaluation value distribution 5. In addition, for the specific production flow of the updated evaluation value distribution 5, the same flow as when the evaluation value distribution 5 is newly created (see FIG. 10) can be used.

＜5. Effect＞ According to this embodiment, the evaluation value of each time series data included in the unit processing data selected by the user is calculated. Then, the evaluation values are statistically normalized to create an evaluation value distribution 5 indicating the distribution of normalized evaluation values. In the case where the evaluation value distribution 5 is created in this way, when time-series data is newly generated by execution of the recipe, based on the evaluation value of the time-series data (specifically, after the standardization of the evaluation value obtained by scoring) Value) to determine abnormality. When the evaluation value distribution 5 is used in the abnormality determination, the evaluation value distribution 5 is updated in this embodiment. Therefore, in the abnormality determination, for example, the latest tendency of the time series data can be considered. As described above, according to the present embodiment, abnormality detection using time series data can be performed with sufficient accuracy in consideration of the passage of time.

<6. Modifications> Hereinafter, a modification related to the update of the evaluation value distribution 5 will be described.

<6.1 First Modification> In the above embodiment, when there is a change in the content of the recipe, the evaluation value distribution 5 is updated. However, the present invention is not limited to this, and the evaluation value distribution 5 may be updated every time scoring is performed.

FIG. 15 is a flowchart showing an outline of the overall processing flow for abnormality detection using time-series data in this modification. In the above-described embodiment, the processing from step S30 to step S50 is repeated until there is a change in the content of any recipe (see FIG. 7 ). On the other hand, in this modification, after determining the abnormality based on the result of the score (step S40) (step S50), it is necessary to update the evaluation value distribution 5 (step S60). In addition, an evaluation value calculation step for abnormality determination is implemented in step S40.

此處，μn+1是評價值分佈5的製作源的單位處理資料的數量增加至n+1個的狀態下的評價值的平均值，μn是評價值分佈5的製作源的單位處理資料的數量為n個的狀態下的評價值的平均值，xn+1是所追加的單位處理資料的評價值，σ2n+1是評價值分佈5的製作源的單位處理資料的數量增加至n+1個的狀態下的評價值的分散，σ2n是評價值分佈5的製作源的單位處理資料的數量為n個的狀態下的評價值的分散。In addition, in order to create the evaluation value distribution 5, it is necessary to calculate the average value and the standard deviation based on the processing data of all the units as the production source. That is, in order to update the evaluation value distribution 5 every time scoring is performed, it is necessary to calculate the average value and the standard deviation every time scoring. Regarding this, suppose that every time the score is calculated using all the unit processing data of the production source of the evaluation value distribution 5 to calculate the average value and the standard deviation, the load for the calculation becomes very large. Therefore, when the number of unit processing data of the production source of the evaluation value distribution 5 is increased from n to n+1, as long as the following formula (2) to formula (4) are used, the average value and dispersion ( The square of the standard deviation).

Here, μn+1 is the average value of the evaluation values in a state where the number of unit processing data of the production source of the evaluation value distribution 5 is increased to n+1, and μn is the unit processing data of the production source of the evaluation value distribution 5 The average value of the evaluation values in the state where the number is n, xn+1 is the evaluation value of the added unit processing data, and σ2n+1 is the number of unit processing data of the source of the evaluation value distribution 5 increased to n+1 Σ2n is the dispersion of the evaluation values in the state where the number of unit processing materials of the source of the evaluation value distribution 5 is n.

When using the above formula (3) to find μn+1, μn is already found, and when using the above formula (4) to find σ2n+1, σ2n is already found. Therefore, the average value and the standard deviation used to create the updated evaluation value distribution 5 can be obtained with a relatively low load (the standard deviation can be easily obtained by dispersion).

If the number of unit processing data of the source of the evaluation value distribution 5 is small, good accuracy cannot be obtained regarding the abnormality determination of the time series data. Regarding this point, according to this modification, the evaluation value distribution 5 is updated every time scoring is performed, so the accuracy of abnormality determination is gradually improved. Moreover, although it takes some time until the average value or standard deviation converges to a value within a fixed range (sufficient accuracy can be obtained regarding abnormality determination), even in a situation where the unit processing data as a result of the recipe execution has not been fully obtained, Various setting operations related to the creation of the score or evaluation value distribution 5 may be performed in advance.

<6.2 Second Modification> In the above embodiment, the evaluation value distribution 5 is created and updated based on the unit processing data arbitrarily selected by the user. However, the present invention is not limited to this, and the evaluation value distribution 5 may be updated based on the unit processing data obtained by the processing in the designated processing unit 222.

FIG. 16 is a flowchart showing the detailed flow of updating the evaluation value distribution 5 in this modification. In this modification, when the evaluation value distribution 5 is updated, first, the scoring result (data of the evaluation value) is extracted (step S600). In step S600, for example, based on an evaluation value distribution 5, the scoring results regarding the 1000 unit processing data obtained recently are extracted.

Next, based on the scoring result extracted in step S600, the deviation (dispersion or standard deviation) of the evaluation value is calculated for each processing unit 222 (step S601). In addition, at this time, the data of the evaluation value is not standardized. In addition, when a distribution (distribution of evaluation values) is made based on the score result extracted in step S600, the distribution is schematically shown in FIG. 17 for example, and is different for each processing unit. Here, it is generally considered that the more processing units 222 included in the output results, the more time series data with a higher degree of abnormality, the greater the deviation based on the distribution. Therefore, as described above, in step S601, the deviation of the evaluation value is calculated for each processing unit 222. Then, the designation of the processing unit 222 that obtains the smallest deviation among the deviations calculated in step S601 is performed (step S602).

Subsequently, for example, from the 1000 unit processing materials obtained recently, the unit processing materials obtained by the processing in the processing unit 222 specified in step S602 are extracted (step S603 ). Next, for each time-series data contained in the unit processing data extracted in step S603 (hereinafter referred to as "extracted unit processing data"), the evaluation value is calculated (step S604), and further, in step Standardization of the evaluation value calculated in S604 (step S605). In addition, in step S605, the normalization of the evaluation value is also performed using the above formula (1). Finally, for each parameter (ie, each time series data), an updated evaluation value distribution 5 is created based on the data of the standardized evaluation value (step S606).

In addition, in this modified example, the deviation calculation step is implemented in step S601, the processing unit designation step is implemented in step S602, the unit processing data extraction step is implemented in step S603, and the evaluation value calculation step for distribution update is implemented in step S604. Steps S605 and S606 implement the evaluation value distribution creation step.

According to this modification, even in the case where it is difficult to select the unit processing material that becomes the production source of the evaluation value distribution 5, the processing deemed stable can be selected (designated) based on the scoring result of each processing unit 222 Unit 222. Then, based on the unit processing data obtained by the processing in the selected processing unit 222, an updated evaluation value distribution 5 is created. Therefore, the abnormality determination using the evaluation value distribution 5 becomes highly accurate. As described above, according to the present modification, even when it is difficult to select the unit processing data that is the source of the evaluation value distribution 5, the evaluation value distribution 5 can be updated so that the abnormality of the time series data can be accurately determined.

In addition, in the above example, the designation of the processing unit 222 in step S602 is performed considering only the deviation of the evaluation value. Regarding this, for example, the occurrence of the following cases (case) is also considered, as shown in FIG. 18, compared with the distribution corresponding to the processing units containing more time series data with a relatively low degree of anomaly, and containing more abnormal The corresponding deviation of the distribution of the processing unit of the time series data with relatively high degree is small. Therefore, for example, in step S601 (see FIG. 16 ), in addition to the deviation of the evaluation value, the average value of the evaluation value may be calculated, and in step S602, both the deviation of the evaluation value and the average value of the evaluation value may be considered. The person specifies the processing unit 222. At this time, the statistical value calculation step is realized through step S601.

<6.3 Third Modification> In the above embodiment, the updated evaluation value distribution 5 is created by the same flow (see FIG. 10) as when the evaluation value distribution 5 is newly created. However, the present invention is not limited to this, and artificial intelligence (Artificial Intelligence, AI) technology may also be used to determine the updated evaluation value distribution 5. Therefore, in the following, a method of using the AI technology for updating the evaluation value distribution 5 will be described with reference to FIGS. 19 to 21. In this modification, it is assumed that the evaluation value distribution 5 common to all processing units 222 is prepared for each parameter. Furthermore, in the following, attention is paid to the evaluation value distribution 5 regarding one parameter.

In this modification, a neural network including, for example, an input layer, an intermediate layer, and an output layer as shown in FIG. 19 is prepared as a learner. The number of layers of the intermediate layer is not limited, and the number of units of a certain layer may be different from the number of units of other layers among a plurality of intermediate layers. The number of cells in the input layer is not particularly limited as long as the data of the evaluation value distribution 5 can be appropriately input. In addition, as the type of neural network, for example, a general forward propagation neural network or a convolutional neural network can be used.

As shown in FIG. 19, data of the evaluation value distribution 5 is input to the input layer, and score data is output from the output layer. For example, when the input layer includes 100 units U(1) to U(100) as shown in FIG. 20, 99 thresholds TH(1) to TH(99) are set to divide the desirable range of evaluation values into one One hundred ranges. In addition, regarding these thresholds TH(1) to TH(99), it is assumed that the relationship of "TH(1)<TH(2)<TH(3)<...<TH(99)" is satisfied. Under this premise, when the data of a certain evaluation value distribution 5 is registered in the input layer, for example, a value less than the threshold TH(1) (as the value of the evaluation value) is input to the first unit U(1) For the number of degrees, enter the number of degrees above the threshold TH(1) and less than the threshold TH(2) for the second unit U(2) (see FIG. 20). When the data of the evaluation value distribution 5 is registered in the input layer as described above, the data is propagated forward based on the weight/bias value in the neural network at this time, and the score data is output from the output layer.

In the case where the neural network described above is prepared as a learner, the neural network must be learned before the evaluation value distribution 5 is actually updated. In this modification, for learning of a neural network, a plurality of learning materials as shown schematically in FIG. 21 are prepared, each including an evaluation value distribution 5 and scores as teaching materials. In addition, in FIG. 21, the material of the part indicated by the symbol 59 corresponds to one learning material. For each evaluation value distribution 5, for example, a score of 0 points or more and 5 points or less is assigned.

At the time of learning, regarding each learning material, the square error between the score obtained by giving the data of the evaluation value distribution 5 to the input layer (the score output from the output layer) and the score used as the teaching material is obtained. In addition, by the error back propagation method, the value of the weight/bias in the neural network is obtained in such a way that the sum of the square errors of all learning materials becomes minimum.

When the evaluation value distribution 5 is updated, a neural network that has been previously learned in the above manner is used. Specifically, using the learned neural network, the updated evaluation value distribution 5 is set by the following procedure.

First, for each processing unit 222, for example, an evaluation value distribution 5 is created using the latest 1000 time series data about the corresponding parameters that have been obtained recently. Thus, if the number of processing units 222 is 12, 12 evaluation value distributions 5 are created. In addition, the data of the 12 evaluation value distributions 5 corresponding to the 12 processing units 222 are sequentially input to the input layer of the neural network. As a result, 12 points of data are sequentially output from the output layer of the neural network. In this modification, the evaluation value distribution 5 that obtains the highest score (the best score) among the 12 scores is defined as the updated evaluation value distribution 5.

As described above, in this modification, a neural network (learner) that has previously learned using a plurality of learning materials each including an evaluation value distribution 5 and scores as teaching materials is input to a plurality of processing units 222 Corresponding multiple evaluation value distributions 5, the evaluation value distribution 5 in which the score output from the neural network is the best among the multiple evaluation value distributions 5 is determined as the updated evaluation value distribution 5.

According to this modification, when the evaluation value distribution 5 is updated, for each processing unit 222, the evaluation value distribution 5 is created based on the latest time series data. In addition, from the plurality of evaluation value distributions 5, the evaluation value distribution 5 determined to be the best by the AI technology is defined as the updated evaluation value distribution 5. Therefore, the abnormality determination using the updated evaluation value distribution 5 becomes highly accurate. As described above, according to the present modification, as in the second modification, even when it is difficult to select the unit processing data as the production source of the evaluation value distribution 5, the evaluation value distribution 5 can be updated so that the high Accurately determine the abnormality of time series data.

＜7. Others＞ The present invention has been described in detail above, but the above description is only illustrative in all aspects and not restrictive. It should be understood that numerous other changes or modifications can be made without departing from the scope of the invention.

5: Evaluation value distribution 11: CPU 12: Main memory 13: auxiliary memory device 14: Display 15: Input section 16: Communication Control Department 17: Recording Media Reading Department 51, 59: Symbol 61: Start time point input box 62: End time point input box 63: processing unit designation box 64: Recipe designation box 65: Extracted data display area 66: OK button 100: data processing device 110: Unit processing data selection department 120: Evaluation value calculation unit 130: Evaluation value distribution production department 140: Evaluation value distribution update section 150: abnormality judgment unit 160: Data storage department 161: Data processing program 162: Time series data DB 163: Benchmark data DB 164: Evaluation value distribution data DB 200: substrate processing device 210: Locator Department 212: substrate container holding part 214: Locator robot 220: Processing Department 222: Processing unit 224: substrate transfer robot 230: substrate interface 300: communication line 400: recording media 500: abnormal judgment object setting screen 600: Unit processing data selection screen 700: Parameter specification screen A～E: parameter L1: Level 1 L2: Level 2 L3: Level 3 L4: Level 4 S10~S60, S110~S113, S600~S606: Steps U(1)～U(100): unit

1 is a block diagram showing the overall configuration of a data processing system (data processing system for a substrate processing apparatus) according to an embodiment of the present invention. 2 is a diagram showing a schematic configuration of a substrate processing apparatus in the above embodiment. FIG. 3 is a diagram showing a certain time series data in the above embodiment. FIG. 4 is a diagram for explaining unit processing data in the embodiment. FIG. 5 is a block diagram showing the hardware configuration of the data processing device in the embodiment. 6 is a diagram for explaining the evaluation value distribution in the above embodiment. FIG. 7 is a flowchart showing an overview of the overall processing flow for abnormality detection using time-series data in the above embodiment. 8 is a diagram showing an example of an abnormality determination object setting screen in the embodiment. FIG. 9 is a diagram for explaining the determination of the degree of abnormality in the embodiment. FIG. 10 is a flowchart showing a detailed flow of creating an evaluation value distribution in the above embodiment. 11 is a diagram showing an example of a unit processing material selection screen in the embodiment. 12 is a diagram showing an example (an example after display) of a parameter designation screen in the embodiment. 13 is a diagram showing an example of a parameter designation screen (an example after a user designates a parameter) in the embodiment. FIG. 14 is a diagram for explaining the update of the evaluation value distribution in the embodiment. FIG. 15 is a flowchart showing the outline of the overall processing flow for abnormality detection using time-series data in the first modification of the embodiment. 16 is a flowchart showing a detailed flow of updating the evaluation value distribution in the second modification of the embodiment. 17 is a diagram for explaining the creation of a distribution of evaluation values for each processing unit in a second modification of the embodiment. FIG. 18 is a diagram for explaining that it is preferable to consider the evaluation value in addition to the deviation in the second modification of the embodiment. FIG. 19 is a diagram showing an example of a neural network prepared as a learner in a third modification of the embodiment. 20 is a diagram for explaining data input to an input layer of a neural network in a third modification of the embodiment. 21 is a diagram for explaining learning materials in a third modification of the embodiment.

S10~S60: The overall processing flow steps of anomaly detection using time series data

Claims (14)

A data processing method that uses multiple time series data obtained by unit processing as unit processing data and processes multiple unit processing data. The data processing method includes: The evaluation value distribution utilizing step performs processing using an evaluation value distribution that represents the degree of each value of the evaluation value obtained by evaluating each time series data; and The evaluation value distribution updating step updates the evaluation value distribution. The data processing method as described in item 1 of the patent application scope, in which The plurality of unit processing data is data obtained by executing recipes by the substrate processing device, When there is a change in the content of the recipe, the evaluation value distribution update step is executed. The data processing method as described in item 2 of the patent application scope, in which The multiple time series data are time series data about multiple parameters, The evaluation value distribution is set for each parameter, In the evaluation value distribution updating step, only the evaluation value distribution corresponding to the parameter whose content is changed is updated. The data processing method as described in item 3 of the patent application scope, in which In the evaluation value distribution update step, the evaluation value distribution corresponding to the parameter added with the change in the content of the recipe is created based on the stored evaluation value data. The data processing method as described in item 3 of the patent application scope, in which In the evaluation value distribution update step, the evaluation value distribution corresponding to the parameter specified from the outside is newly created. The data processing method as described in item 1 of the patent application scope, in which The step of using the evaluation value distribution includes: An evaluation value calculation step for determining the abnormality, calculating the evaluation value for the time series data contained in the newly acquired unit processing data; and The abnormality determination step performs the abnormality of the time series data included in the newly obtained unit processing data based on the evaluation value distribution and the evaluation value calculated in the evaluation value calculation step for abnormality determination determination. The data processing method as described in item 6 of the patent application scope, in which Whenever the evaluation value calculation step for abnormality determination is executed, the evaluation value distribution update step is executed. The data processing method as described in item 1 of the patent application scope, in which The unit processing is processing performed by a substrate processing apparatus having a plurality of processing units for one substrate as a recipe, The step of updating the evaluation value distribution includes: The deviation calculation step calculates the deviation of the evaluation value for each processing unit based on the evaluation value of each time series data; A processing unit specification step, specifying a processing unit that obtains the smallest deviation among the deviations calculated in the deviation calculation step; A unit processing data extraction step, extracting unit processing data corresponding to the processing unit specified in the processing unit specifying step from the plurality of unit processing data; An evaluation value calculation step for distribution update, calculating an evaluation value for each time series data contained in the extracted unit processing data, the extracted unit processing data being the unit processing data extracted in the unit processing data extraction step; as well as The evaluation value distribution creation step creates an updated evaluation value distribution for each type of time series data based on the evaluation values for each time series data calculated in the distribution update evaluation value calculation step. The data processing method as described in item 1 of the patent application scope, in which The unit processing is processing performed by a substrate processing apparatus having a plurality of processing units for one substrate as a recipe, The step of updating the evaluation value distribution includes: Statistical value calculation step, based on the evaluation value of each time series data, calculate the average value and deviation of the evaluation value for each processing unit; Processing unit designation step, designating the processing unit considering the average value and deviation calculated in the statistical value calculation step; A unit processing data extraction step, extracting unit processing data corresponding to the processing unit specified in the processing unit specifying step from the plurality of unit processing data; An evaluation value calculation step for distribution update, calculating an evaluation value for each time series data contained in the extracted unit processing data, the extracted unit processing data being the unit processing data extracted in the unit processing data extraction step; as well as The evaluation value distribution creation step creates an updated evaluation value distribution for each type of time series data based on the evaluation values for each time series data calculated in the distribution update evaluation value calculation step. The data processing method as described in item 1 of the patent application scope, in which The unit processing is processing performed by a substrate processing apparatus having a plurality of processing units for one substrate as a recipe, In the evaluation value distribution updating step, a plurality of evaluation value distributions corresponding to the plurality of processing units are input to the learner, and the score output from the learner among the plurality of evaluation value distributions is the most The preferred evaluation value distribution is determined to be the updated evaluation value distribution, and the learner performs learning in advance using a plurality of learning materials each including an evaluation value distribution and a score as teaching materials. The data processing method as described in item 10 of the patent application scope, in which The evaluation value distribution input to the learner in the evaluation value distribution updating step is an evaluation value distribution created based on the evaluation values of the time series data obtained by the execution of the most recent recipe. The data processing method as described in item 10 of the patent application scope, in which The learner is a neural network with an input layer including multiple units, an intermediate layer including multiple units, and an output layer including one unit, Each unit of the input layer is associated with a range obtained by dividing a range of values that can be obtained as evaluation values with a predetermined threshold, For each cell of the input layer, input the degree of the value contained in the range associated with each cell. A data processing device takes multiple time series data obtained by unit processing as unit processing data and processes multiple unit processing data. The data processing device includes: The evaluation value distribution utilization section performs processing using an evaluation value distribution that represents the degree of each value of the evaluation value obtained by evaluating each time series data; and The evaluation value distribution update unit updates the evaluation value distribution. A computer-readable recording medium that stores a data processing program for causing a computer to perform an evaluation value distribution utilization step and an evaluation value distribution update step. The computer will process various time series obtained by unit processing Use the data as a unit to process data, and process data from multiple units: The evaluation value distribution uses a step to perform a process using an evaluation value distribution, the evaluation value distribution representing the degree of each value of the evaluation value obtained by evaluating each time series data; and The evaluation value distribution update step updates the evaluation value distribution.

Images